Introducing the Opportunity for Commoditizing Intelligence
The world looks at compute as plumbing and overhead: an expense line-item or the intangible output of data centers. Instead, compute should be seen as an asset like oil, directly accessible in liquid markets and built into sophisticated financial instruments.
Markets around critical resources begin to financialize once they reach a size where the pain of not having a hedge exceeds the friction of building one, or when insatiable demand for speculation and exposure requires the creation of novel assets and instruments. And that’s exactly where AI compute finds itself today.
AI infrastructure capex has exploded into the hundreds of billions as neo-clouds pledge GPU fleets and customer contracts as collateral for multi-billion-dollar loans, and startups reserve capacity months in advance as an existential requirement despite uncertainty around end-demand or funding ability.
Across the AI stack, buyers and sellers are being forced to make long-duration financial decisions around an asset with inherent uncertainty around shelf-life and value. At the same time, investors are screaming into the void for an option to get exposure beyond a dozen megacap public stocks or opaque SPVs into the large labs.
The point here isn’t just that compute has become expensive, but that it’s become so volatile, valuable, and strategically necessary that it finally needs a market. This is the natural evolution for any commodity as it evolves from a raw resource into a financial instrument.
But compute is more complicated than commodities that have come before it.
The biggest hurdle is that GPUs aren’t inherently fungible. In traditional commodity markets, fungibility is what allows you to mark a barrel of oil as interchangeable with any other barrel of the same grade, and therefore trade it freely. For example, crude oil can be graded and standardized via attributes like sulfur content and density.
But unlike crude oil, compute is not a naturally homogenous asset. GPUs vary by chip generation, cluster topology, interconnect speeds, geography, latency, SLA terms, uptime, and utilization. Two clusters with the same number of chips can have drastically different economic value depending on where they are, how they’re connected, who’s running them, and what kinds of workloads they can support.
This is the core market design problem we’re facing today: how do you write a fungible contract on top of a fundamentally nonfungible physical asset?
The solution will involve novel approaches to normalization and index calculations.
Compute markets will need standardized benchmarks that incorporate granular data flows, indices, and qualitative ratings. New contract specifications are also needed to turn messy physical differences into something legible enough to trade or reliably plan corporate strategy around.
But standardizing compute as a unit of account will still only be the first step. The next question is how those contracts should settle. The answer may differ depending on the market participant in question.
Cash-settled GPU futures settle via USD payout based on an index of GPU rental prices. Cash-settled contracts are more liquid, with lower friction, making them especially attractive to financial institutions, macro hedge funds, and energy trading desks who want speculative exposure to compute prices but don’t want the logistical burden of touching any physical assets.
On the other hand, physically-settled contracts require the seller to deliver live access to a designated GPU cluster under the specific terms of the contract to the buyer. This structure is best suited for those who actually need compute, such as startups, cloud infrastructure providers, and enterprise buyers. If you’re planning an enormous training run or need guaranteed capacity for a spiky real-time consumer application, a cash payout when prices fluctuate or hardware suddenly becomes scarce does not provide you with the resource you actually need. Conversely, physical platforms and operators often use this structure to dynamically offload underutilized inventory.
This distinction is important to call out, as different market participants are solving for different objectives.
Some are securing control or hedging against the physical asset, while others are looking to make directional bets from an investment perspective. A hedge fund may only need price exposure, while a frontier lab or enterprise likely needs actual physical capacity. That means GPU markets probably won’t converge around one perfect contract but will instead evolve into diverse instruments addressing different kinds of risks and requirements.
Besides the fact that physically settled contracts are especially high friction, they also present a unique version of a familiar commodity problem referred to as vintage drift. Physical delivery of a depreciating asset whose generation goes obsolete in 12-18 months is a logistical and economic nightmare that other commodities like crude oil don’t have to deal with.
And that’s why cash-settled perpetual futures (perps) built around blockchains and stablecoin rails may be the cleaner initial wedge for GPU markets.
Traditional futures force traders to roll over contracts as they expire, exposing them to the extreme vintage drift of the AI hardware lifecycle where a generation of chips rapidly shifts from premium to discounted. Perps avoid this expiration loop entirely, allowing traders to hold long-term structural hedges on the direction of compute costs without managing physical delivery dates. They may not be a perfect solution, but they do allow for a market to financialize compute without forcing it to solve for physical delivery on day one.
As a final point, all of this makes the venue question very important.
If compute is a 24/7, global, dollarized asset, then the market around it needs to match those features. A stablecoin-settled contract that trades 24/7 feels more at home in that world than a contract that clears in one time zone only during market hours. For cash-settled perps, decentralized oracles could be leveraged to stream physical market indices onchain, and smart contracts could calculate funding rates and execute liquidations autonomously.
Platforms like Architect are already showing the world what this could look like by launching exchange-traded perps on GPU and RAM pricing through AX. In parallel, teams like Ritual are creating an onchain environment where AI agents natively discover, hedge, and purchase their own compute. And Odyn is leading the buildout of a globally distributed compute substrate that can eventually power physical settlement as well.
There are still open questions and design challenges around scaling and monetizing compute, but it’s clear that AI infrastructure is evolving beyond a pure cost center into a financial asset class with its own indexes, markets, and financial products. It has to.
Compute is becoming the first true commodity market for intelligence, and the decisions behind when and how companies lock in compute capacity are becoming crucial treasury and investment strategies.
Teams that develop resilient compute strategies will outperform those who treat it as simple plumbing infrastructure.
While the reflexive instinct in AI is that the opportunity is always the next model, the next chip, or the next capability, the real lesson is that during any supercycle, the largest and most durable opportunities accrue not to the thing itself but to the market structure that forms around it.
